Economic Policies vs. Identity Politics - The Rise of a Right-wing, Nationalist Party in India Robustness Check

3.2.1 Robustness Check: Disaggregated Welfare Benefits In the preceding section, we aggregated the different kinds of state/central benefits that a household received to generate a single state/central benefit count variable. This implicitly weighted different kinds of state/central benefits equally, but of course different benefits may of different value. In fact, as we saw above, over the period 2013–2018 there were important changes in the composition of state benefits: households became less likely to receive NREGS workfare benefits, but they became more likely to hold BPL cards, have toilets in the neighourhood, have subsidised housing or participate in microfinance (“self-help”) groups. If NREGS benefits are more influential than other programs in determining which party households vote for, then our results above may have overestimated the effectiveness of benefits in determining the TMC’s vote share. In Table 5, we estimate the regressions separately for NREGS, BPL cards, drinking water access, in-house toilets, bank accounts (Jandhan) and cooking gas subsidies (Ujjwala). The IV estimates use “leave-one-out” instruments constructed separately for each benefit program. The first stage regressions have high F-statistics for each program and round. As we see in the table, the coefficients vary considerably across programs: NREGS has a much larger coefficient than the other two state benefits, and is the only state benefit with a statistically significant IV effect of 0.19 for the 2018 round (significant at 5%). Among the central benefits, only the Jandhan program has a significant effect: it increases the likelihood that a household voted for the BJP by 0.14 (significant at 10%). The NREGS was more effective at boosting the TMC vote share in 2013 (coefficient 0.22, also significant at 5%), but it also had a strong negative (-0.13) impact on the BJP’s vote share in 2018. Interestingly, while Deshpande et al. (2019) argue that in 2019, voters were more likely to credit the BJP for the NREGS despite it being a legacy of the previous UPA government, the NREGS beneficiaries are less likely to vote for the BJP. The vote of the NREGS beneficiaries is dispersed among regional parties despite the observed centralization of attribution of credit in the 2019 elections. Consistent with this argument, our results show that the voters clearly credit the regional incumbent party (TMC) for this program.19 In Table 6 we compute the implied combined effect on predicted vote shares. If we focus on the NREGS and Jandhan programs alone (the only benefits that have statistically significant impacts on votes), then the changes in the receipt of benefits predict that the BJP’s vote share would fall by 5.4%. This is the net effect of an increase of 2.7% points caused by the flow of Jandhan benefits, and a decrease of 8.1% (5.9 + 2.2) due to the decrease in NREGS benefits. At the same time, the TMC’s vote share is predicted to fall by 4.9% (a decrease of 3.2% = 11.8 − 8.6) due to reduced NREGS benefits and an additional decrease of 1.7% due to Jandhan benefits. If we aggregate the impacts of all six programs, then the benefit flows predict that the TMC’s vote share would increase by 1.1%, while the BJP’s vote share would decrease by 4.3%. These 19 We have categorized NREGS as a state benefit, but the results do not depend on this categorization. Table 5: Effect of Major State and Central Benefits on Voting Vote TMC Vote BJP First Stage F OLS (1) IV (2) OLS (3) IV (4) 0.035 (0.032) -0.001 (0.031) 0.038 (0.033) 0.094 (0.104) -0.093 (0.080) -0.010 (0.110) 0.040 (0.033) 0.025 (0.022) -0.025 (0.028) 0.013 (0.102) 0.142* (0.076) -0.041 (0.103) 0.102*** (0.029) 0.023 (0.025) 0.076* (0.040) 3,507 0.044 0.188** (0.080) 0.099 (0.087) 0.092 (0.077) 3,507 0.025 -0.045* (0.024) -0.019 (0.024) -0.089** (0.037) 3,507 0.059 -0.130* (0.076) -0.128 (0.084) -0.095 (0.080) 3,507 0.024 0.119*** (0.020) 0.225*** (0.047) 0.010 (0.008) 0.042* (0.022) 761.81 BPL Card -0.025 (0.047) 0.011 (0.160) 0.036 (0.032) -0.093 (0.080) 175.64 Drinking Water (Panchayat) -0.012 (0.031) -0.021 (0.052) -0.002 (0.008) -0.002 (0.015) 436.60 3,507 0.379 3,507 0.368 3,507 0.061 3,507 0.044 (5) Voting in 2018 Central Benefits 2014 – 2018 In House Toilet Jandhan Ujjwala 343.27 647.30 197.41 State Benefits 2014 – 2018 NREGS BPL Card Drinking Water (Panchayat) Number of Observations R-squared 240.15 150.41 856.96 Voting in 2013 State Benefits 2010–2013 NREGS Number of Observations R-squared Notes: Dependent variable voting for TMC (Columns 1 and 2) and BJP (Columns 3 and 4). IV regressions use the Leave-out instrument. Regressions also control for religion, caste and migration status of household, gender, marital status, education and occupation of household head, household size, change in employment income and farm value added over the period, whether the household is a swing voter and whether household resides in Hugli. Standard errors, clustered at the Village level in parentheses. Significance: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗ p < 0.1. 23 Table 6: Predicted Effect of Major Central and State Benefits on Voting Mean (Q) (1) Unit Effect on Vote Share b TMC (βTb M C ) BJP (βBJP ) (2) (3) Total Effect on Vote Share t,b TMC (Q × βTt,bM C ) BJP (Q × βBJP ) (4) (5) Central Benefits 2014–2018 In House Toilets Jandhan Ujjwala 0.155 0.187 0.161 0.094 -0.093 -0.010 0.013 0.142 -0.041 0.015 -0.017 -0.002 0.002 0.027 -0.007 0.455 0.374 0.259 0.188 0.099 0.092 -0.13 -0.128 -0.095 0.086 0.037 0.024 -0.059 -0.048 -0.025 0.523 0.036 0.303 0.225 0.011 -0.021 0.042 -0.093 -0.002 0.118 0.000 -0.006 0.022 -0.003 -0.001 Central Benefits State Benefits -0.001 0.012 0.007 -0.050 Total Predicted Effect of Benefits (Central + State) Actual Change 0.011 -0.05 -0.043 0.18 State Benefits 2014–2018 NREGS BPL Card Drinking Water (Panchayat) State Benefits 2010–2013 NREGS BPL Card Drinking Water (Panchayat) Predicted Effect Notes: Predicted Change in vote share for party p (Πp ) is given by state or central benefits. No central benefits in 2013. P b (Q b β 2018,b p − Qb βp2013,b ) aggregated over are conservative estimates, since we do not consider the role of new benefits or the increased delivery of benefits from other smaller state schemes such as the conditional cash transfer to young girls for delaying marriage (Kanyashree), self-help groups and medical help provided by the local government. Including them would predict that the TMC’s vote share would increase even further, and the BJP’s vote share would decrease even further. Clearly, incorporating the composition of the benefit programs does not change our inference that benefits alone cannot explain the substantial increase in the BJP’s vote share over the period 2014 to 2019. 3.2.2 Robustness Check: Targeting Patterns and Heterogenous Voting Impacts The preceding analysis ignored the fact that households of different economic status may respond differently to the benefits. For instance, if poorer households are more likely to change whom they vote for in response to benefits, then delivering benefits to the poor would be more effective at raising the vote share of the incumbent party. If central benefits were better targeted to the poor than state benefits, then predictions based on aggregate benefit flows would underestimate 24 Figure 6: Progressivity of Benefits: Number of Central and State Benefits Received by Predicted Income Quantiles Notes: Predicted Income Quantiles computed using the predicted value of income from regression results presented in Table A1. the predicted impacts on the BJP vote share. Alternatively, if state benefit schemes became less well-targeted to the poor after 2013, then even if on average the benefit flows increased, the poor would not receive the same increases in benefit as the non-poor, and the aggregate analysis predict a larger impact on the TMC’s share than we actually observe. Figure 6 shows the targeting patterns of state and central benefits across four income quartile groups. Note first that the distribution of state benefits became markedly progressive after 2013 relative to the 2010–13 and after 2013, state benefits were also more progressively distributed than central benefits. Hence we do not see evidence that central benefits were better targeted to the poor than state benefits, either before or after 2013. Table 7 shows OLS estimates of voting impacts of state and central benefits from regressions separately for the four different income quartiles using the same specification as in equation (4). Only voters in the second quartile responded significantly to receiving central benefits, by switching their vote to the BJP. However, all four quartiles responded to state benefits after 2013 25 Table 7: Impact of Benefits on Votes for Different Income Quartiles Predicted Income Quantile 1 2 Central Benefits 2014–2018 3 4 1 2 State Benefits 2014–2018 3 4 Voting 2018 TMC BJP Left TMC BJP Left (1) (2) (3) (4) (5) (6) 0.007 (0.022) -0.029 (0.030) 0.006 (0.046) 0.077** (0.030) 0.011 (0.019) 0.057** (0.025) 0.037 (0.033) -0.005 (0.020) -0.003 (0.013) -0.006 (0.013) 0.015 (0.027) -0.013 (0.015) 0.057*** (0.013) 0.064*** (0.015) 0.039* (0.023) 0.081*** (0.020) -0.049*** (0.011) -0.058*** (0.014) -0.014 (0.018) -0.047*** (0.017) -0.007 (0.007) -0.003 (0.006) -0.035** (0.014) -0.009 (0.008) 0.012 (0.014) 0.028** (0.014) 0.018 (0.026) 0.025 (0.018) 0.002 (0.006) 0.011** (0.005) 0.004 (0.012) -0.005 (0.009) -0.028* (0.014) -0.045*** (0.014) -0.017 (0.030) -0.033** (0.016) 1 2 State Benefits 2010–2013 Voting 2013 3 4 Notes: In columns 7–10, dependent variable is switching to or against a particular party in 2018 (relative to 2013). OLS regressions presented. Regressions also control for religion, caste and migration status of household, gender, marital status, education and occupation of household head, household size, whether the household is a swing voter and whether household resides in Hugli. Predicted Income Quartiles computed using the predicted value of income from regression results presented in Table A1. Standard errors, clustered at the Village level in parentheses. Significance: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗ p < 0.1. 26 Table 8: Targeting of Benefits and Predicted Impact on Votes Predicted Income Quartile Mean (Q) (1) Unit Effect on Vote Share t TMC βTt M C,j BJP βBJP,j (2) (3) Total Effect on Vote Share t,b TMC (Q × βTt,bM C,j ) BJP (Q × βBJP,j ) (4) (5) Central Benefits 2014–2018 1 2 3 4 0.681 0.527 0.491 0.334 0.00674 -0.0292 0.00561 0.0766 0.0109 0.0573 0.0372 -0.00513 0.0046 -0.0154 0.0028 0.0256 0.0074 0.0302 0.0183 -0.0017 State Benefits 2014–2018 1 2 3 4 2.349 1.890 1.548 0.906 0.0575 0.0643 0.0391 0.0807 -0.0491 -0.0575 -0.0144 -0.0475 0.1351 0.1215 0.0605 0.0731 -0.1153 -0.1087 -0.0223 -0.0430 State Benefits 2010–2013 1 2 3 4 1.345 1.188 1.009 0.936 0.0122 0.0281 0.0176 0.0251 0.00174 0.0114 0.00410 -0.00475 0.0164 0.0334 0.0178 0.0235 0.0023 0.0135 0.0041 -0.0044 Predicted Effect on Change in Vote Shares 1 P b2018 ) Central Benefits ( n j (Q × βp,j 1 P 1 P b2018 b2013 ) State Benefits ( n j Q × βp,j − n j Q × βp,j 0.0044 0.0135 0.0748 -0.0762 Total Predicted effect of Benefits (Central + State) 0.0792 -0.0627 -0.05 0.18 Actual Change Notes: j denotes predicted income quartile (computed using the predicted value of income from regression results presented in Table A1); b denotes the benefit type (state/central); t denotes year (2013, 2018). by roughly the same extent.20 Consistent with the earlier aggregate results, voters in all groups were markedly more responsive to state benefits after 2013. Once again, we find that state benefits were more effective at generating votes for the TMC after 2013. Table 8 shows the implication of these estimates for the vote shares of the two parties. The TMC’s vote share is predicted to increase by 7.9%, while the BJP’s vote share is predicted to fall by 6.2%. The central benefits alone predict a rise in the BJP vote share by 1.35%, which is only 7.5% of the total increase in the vote share of the BJP. Over and above responding to whether they themselves receive a government benefit, voters may also be influenced by perceptions about the fairness of benefit distribution. For example, a household that believes it deserved to receive a benefit but in fact did not receive it, may respond differently if it believed that others who did not deserve it had received it. Critics of the TMC sometimes argue that it “appeases” minorities such as Muslims and immigrants (who are argued to be non-deserving of these benefits) in order to obtain their votes. To the extent that Hindu natives perceive this to be the case, when benefits are distributed to these groups it may cause native Hindus to switch their votes to the BJP. To examine this, in Table 9 we show how state benefits were distributed across different socio-economic groups. Each column 20 The exception is the second quartile, where the point estimate was lower but still remained within the confidence interval for other groups. shows OLS regression coefficients of the likelihood that a specific household received state or central benefits, from a regression on household characteristics in different time periods, with and without village fixed effects. The coefficients on dwelling type and landowning category show the same progressive pattern in the distribution of state benefits that we saw earlier in Figure 6: the state benefit distribution became more progressive after 2013, and was more progressive than the distribution of central benefits after 2013. Controlling for these markers of economic status, we see that non-Hindus were less rather than more likely than Hindus to receive state benefits, that this disproportionate receipt of benefits by Hindus intensified markedly after 2013, and that non-Hindus were also less likely than Hindus to receive central benefits, but this skewness was less pronounced among central benefits than post-2013 state benefits. We also see a similar pattern in that scheduled tribe households were less likely to receive state benefits. Thus we fail to find any evidence of that the TMC administration selectively distributed benefits to minorities; instead, we find that minorities were less likely to receive state benefits. 3.3 The Role of Economic Distress As we have argued in the introduction, there is a large and growing literature from both developed and developing countries on economic insecurity and distress (arising from globalization, tradeshocks or immigration) leading to increased vote-shares of populist and right-wing nationalist parties (see, for example, Colantone and Stanig, 2018, 2019, Autor et al., 2020, Oliveira, 2022). In view of this literature, we examine whether households may have changed whom they voted for in response to their economic situation. For example, if households experienced economic distress after 2013 and held the TMC’s policies responsible for this distress or the lack of government aid in response to the distress, they may have switched their political support from the TMC to the BJP. To measure households’ economic situation, we use two measures: their employment income, and their value-added from the cultivation of potatoes. Recall from Table 1, real employment income fell during the period for 68.75% of households while real potato value-added fell for 8.5% of households. Table 10 shows OLS regression coefficients on the change in potato value added and employment income (Panel A) and whether or not the household experienced a decline in potato value-added and employment income (Panel B) on the likelihood that the household switched its vote away or towards the TMC, and towards BJP, after controlling for benefits received and fixed household characteristics. The regression coefficients are quantitatively negligible and statistically insignificant. Hence we fail to find any evidence that the switch to the BJP was the result of economic distress under the TMC regime. Table 9: Targeting of State and Central Benefits State Benefits 2010–2013 Male Headed Household Non Hindu Household Recent Immigrant Household Non Hindu × Recent Migrant SC Household ST Household Kuchha House Landless Land holding 0 – 0.5 acres Land holding 0.5 – 1 acres Constant Number of Observations R-squared Central Benefits 2014–2018 2014–2018 OLS (1) VFE (2) OLS (3) VFE (4) OLS (5) VFE (6) -0.137 (0.104) -0.120 (0.121) -0.183 (0.148) -0.315 (0.204) 0.072 (0.165) 0.005 (0.080) 0.111* (0.058) 0.331*** (0.071) 0.208*** (0.059) 0.069 (0.045) 1.038*** (0.119) -0.203** (0.095) -0.077 (0.120) -0.111 (0.126) -0.176 (0.172) 0.215 (0.131) 0.006 (0.055) 0.091** (0.037) 0.320*** (0.055) 0.257*** (0.041) 0.084** (0.041) 0.896*** (0.084) -0.131 (0.120) -0.721*** (0.177) 0.043 (0.207) -0.035 (0.455) -0.025 (0.267) -0.343** (0.141) 0.402*** (0.082) 0.695*** (0.115) 0.645*** (0.106) 0.211** (0.080) 1.576*** (0.186) -0.296** (0.133) -0.398*** (0.127) 0.090 (0.261) -0.721 (0.456) 0.229 (0.144) -0.173 (0.132) 0.284*** (0.050) 0.855*** (0.085) 0.673*** (0.067) 0.267*** (0.061) 1.153*** (0.094) -0.010 (0.066) -0.094 (0.074) 0.086 (0.117) -0.010 (0.248) 0.062 (0.104) -0.046 (0.062) 0.179*** (0.035) 0.224*** (0.049) 0.200*** (0.040) 0.096*** (0.031) 0.109* (0.065) 0.030 (0.065) -0.150* (0.080) 0.063 (0.088) -0.132 (0.249) 0.036 (0.082) -0.036 (0.057) 0.152*** (0.031) 0.230*** (0.047) 0.215*** (0.035) 0.090*** (0.032) 0.246*** (0.058) 3,507 0.078 3,507 0.309 3,507 0.244 3,507 0.503 3,507 0.079 3,507 0.226 Notes: OLS regression results presented. Columns marked VFE include village fixed effects. Regressions also control for household size and characteristics of the household head (marital status, educational attainment and occupation, swing voter, Hugli residence). Recent migrant denotes households where the head was not born in the village but migrated less than 10 years previously. Significance: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗ p < 0.1. 3.4 Identity Politics: Variation of Switching Likelihoods across Household Characteristics Having failed to find any evidence of the role of welfare benefits or economic distress in explaining the rise in the BJP vote share, we now consider how this rise varied across different socio-economic groups, with or without controls for benefits or economic distress. Figure 7 shows changes in vote shares across different population sub-groups in the raw data. The top left-hand panel shows that the rise in the BJP vote share was nearly uniform across different landowning categories, rising from 4–6% in 2013 to between 20–24% in 2018. A similar pattern emerges across different income quartiles in the bottom right panel. The top-right panel shows the shift was markedly smaller among non-Hindus (a rise of 6–7%), compared to a rise of 20% among Hindus and 16% among Table 10: Post-2013 Economic Distress and VoteSwitching Likelihood Against TMC (1) To TMC (2) To BJP (3) -0.004 (0.028) -0.001 (0.004) 0.004 (0.018) -0.002 (0.003) -0.001 (0.030) 0.004 (0.004) 3,502 0.090 3,502 0.159 3,502 0.060 0.027 (0.027) 0.008 (0.016) 0.021 (0.024) -0.021 (0.014) - 0.005 (0.024) 0.003 (0.019) 3,505 0.091 3,505 0.160 3,505 0.060 Panel A: Change in Potato value-added Change in Employment Income Number of Observations R-squared Panel B: Decline in Potato value-added Decline in Employment Income Number of Observations R-squared Notes: OLS regression results presented. Regressions also control for characteristics of the household head (gender, education, age, occupation), household size, landholding, identity of household (gender of head, caste, religion, migrant status) change in state benefits received (2014– 2018 — 2010–2013), central benefits received (2014–2018) and Hugli resident. Employment income and Potato Value added in real terms (using the All India Price Index Number for Agricultural Labourer, General Index, 1986 − 87 = 100). Employment Income and Potato value added in Rs. (’0000). Change in employment income = Employment income (2018) – Employment income (2010–2013). Change in potato valueadded = Potato value-added (2018) – Potato value-added (2010–2013). Decline in Employment income = Change in Employment income < 0. Decline in Potato value-added = Change in Potato value-added < 0. STs. The bottom-left panel compares shifts across groups defined by religion and immigration status of the household. We see that among Hindus the rise in BJP vote share did not vary with migrant status: both increased by 20%. However among non-Hindus, it was markedly lower for recent immigrants compared with natives. Hence non-Hindus, and in particular non-Hindu recent immigrants exhibited a much smaller likelihood of switching to the BJP. Strikingly, this is despite the greater discrimination they faced relative to Hindu households in the distribution of state benefits. This indicates there must have been ‘other’ factors that made them reluctant to switch to the BJP — in line with the hypothesis of rising importance of identity politics based on religion and immigration. These results suggest that similar results will obtain after controlling for benefits received: nonHindus would be even more inclined to vote for the TMC if they were to receive the same Figure 7: Vote shares by Household Characteristics. Survey Data Panel A: Land Category Panel B: Religion and Caste Panel C: Immigrant Status Panel D: Predicted Income Quantile Notes: Percentage of Households voting for Left, BJP and TMC in strawpoll conducted in 2013 and 2018 by household characteristics presented. Percentage of households that belong to each land category: 20, 40, 23, 18 (Panel A). Percentage of households that belong to each religion and caste category: 15, 69, 6, 10 (Panel B) Percent of households that belong to each migrant category: 80; 5; 15; 1 (Panel C). Predicted Income Quantiles (Panel D) computed using the predicted value of income from regression results presented in Table A1. state benefits as Hindus. This is verified in Table 11 shows the same results hold for the voteswitching likelihood when we control for other household characteristics, changes in state benefits (post-pre2013 difference) and post-2013 central benefits. Holding these controls fixed, non-Hindu households were 12% less likely to switch away from the TMC, 4.6% more likely to switch towards the TMC, and 16.5% less likely to switch to the BJP (this is heightened further among non-Hindu immigrants). ST households, the other minority that was discriminated against in the distribution of state benefits, exhibited qualitatively similar patterns: 6% less likely to switch against TMC or towards BJP. Table 11: Identity and Voting Patterns: Regression of Vote-Switching Likelihood on Household Characteristics Against TMC (1) To TMC (2) To BJP (3) 0.031 (0.040) 0.038 (0.046) -0.061* (0.036) -0.119*** (0.024) 0.081 (0.095) -0.046 (0.192) 0.215*** (0.047) -0.055 (0.041) -0.055 (0.046) 0.060* (0.032) 0.046* (0.026) -0.007 (0.077) -0.175 (0.146) 0.309*** (0.040) 0.042 (0.039) -0.008 (0.058) -0.066 (0.046) -0.165*** (0.027) 0.086 (0.086) -0.197** (0.092) 0.272*** (0.054) Male Headed Household SC Household ST Household Non Hindu Household Recent Immigrant Household Non Hindu × Recent Immigrant Constant Notes: OLS regression results presented. Regressions also control for characteristics of the household head (gender, education, age, occupation), household size, landholding, change in state benefits received (2014–2018 — 2010–2013), central benefits received (2014–2018) and Hugli resident. Recent immigrant denotes households where the head was not born in the village but migrated less than 10 years previously. 3.5 2018 Panchayat Election Violence During the 2018 village council elections in West Bengal, the media reported a number of violent incidents around the state. Observers suggested that the violence was initiated or perpretrated by supporters of the Trinamool Congress, with the aim of intimidating their political opponents into withdrawing their candidacy. In the subsequent 2019 national elections, some suggested that voters had switched their votes away from the Trinamool Congress in order to punish them for this violence. If this is the case, then we should see that the BJP received more votes in polling booths that fall within the village councils where TMC candidates ran unopposed in the 2018 village council elections. However, as we see in Figure 8, across the 42 parliamentary constituencies in the state, there is no systematic pattern suggesting this was the case. 4 Discussion and Conclusion Various explanations have been provided for the sharp rise in BJP support during the period 2014– 2019. Jaffrelot (2019), Chhibber and Verma (2019) and Choudhary et al. (2020) all emphasize the role of identity politics and that in 2019, caste and religion were the key determinants of voting Figure 8: Unopposed Seats in 2018 Village Council Elections and 2019 Vote shares Panel A: District Level Vote Shares 2019 Change in Vote Shares (2019–2014) Panel B: Assembly Constituency Level Vote Shares 2019 Change in Vote Shares (2019–2014) Notes: Local weighted polynomial regressions and 90% confidence interval presented. Data includes all parliamentary election constituencies in West Bengal. for BJP. This is in contrast to the “issue-based electoral chord” that BJP was able to strike with the voters in 2014. In contrast Deshpande et al. (2019) argue that the rise in BJP support was driven partly by changes in scale of benefits both old and new, and of the effectiveness of the former in raising votes of regional incumbents owing to a greater willingness of voters to shift credit to the BJP government. Accordingly they emphasize that the role of identity politics in explaining BJP’s rising popularity has been exaggerated. Using rich household level survey data from West Bengal, we examine the relative importance of economic factors and identity politics for the increased voter support for BJP between 2014 and 2019. The context is interesting because until the 2019, BJP was not viewed as a major electoral contender in the state and the rise in their vote share was drastic and sudden. We summarize our main findings as follows. First, the sharp increase in the vote share of the BJP cannot be explained by changing structure of state and central government welfare benefits, declining effectiveness of state benefits in generating votes for the TMC, or superior targeting of central benefits. Our results for West Bengal agree with the specific findings of Deshpande et al. (2019) at the all-India level regarding declining scale and effectiveness of NREGS benefits in generating higher votes for regional incumbents after 2014, and the positive impact of the new central benefits in raising the BJP vote share. But our overall conclusion differs markedly, because we additionally incorporate the role of other state programs whose scale and effectiveness in generating TMC votes grew markedly after 2014. Moreover the quantitative magnitude of the positive impact of central benefits on the BJP share was weak in comparison. Hence when assessing the combined impact of changes in different state and central programs, the net changes in vote share were opposite in nature to what was actually observed. We also failed to find any evidence of the role of economic distress or rising anti-TMC incumbency related to violence in the 2018 local government elections. In contrast to the role of welfare benefits or economic distress, we find evidence consistent with the role of shifting identity politics as an important driver of increased support for BJP. Voters belonging to different social identity groups were differentially likely to switch their votes in favor of the BJP, with non-Hindus and recent immigrants significantly less likely to switch their support to the BJP in 2019. Our results are consistent with a number of other recent articles on West Bengal politics that draw attention to political strategies of the TMC. Dey and Sen (2020), Shenoy and Zimmerman (2020), Mahadevan and Shenoy (2023) provide detailed empirical evidence of political clientelism in West Bengal during the post-2011 period after TMC acquired majority control of state and local governments. Bhattacharya and Dasgupta (2023) suggest that the structural characteristics of the rural labour market in West Bengal, and the low labour market participation rates of women in the state create conditions for the continuing prevalence of a system where the electorate is dependent on transfers from the state government that stabilize household incomes in an environment of fluctuating and uncertain earnings. Nath (2022) provides data and ethnographic evidence that since 2011 the TMC has increasingly encouraged religious and cultural festivals and the use of traditional community-based mechanisms of dispute settlement, which encouraged minorities to vote for the TMC. Data limitations prevented us from examining other possible sources of anti-TMC-incumbency arising from citizens assigning TMC leaders and party workers for corruption or governance failures on other dimensions such as security or dispute settlement. They also prevented us from including other districts or states in the analysis. This indicates the need to collect or utilize more data to extend our approach to incorporate other locations and dimensions of governance.